<h1 id="ev-ganin-style-(krea2-style-lora)-v1">Ev Ganin Style (Krea2 Style LoRA) - V1</h1>If you love the Midjourney aesthetic, you have likely seen this distinct style. Ev Ganin is well-known for generating incredibly striking, bold line art combined with highly stylized, exaggerated cartoony proportions. I wanted to take that highly restricted, closed-ecosystem look and bring it natively to Krea 2 so the open-source community could have full, uninhibited control over it.<h3 id="v2-update:-gender-overfit-fixed-and-new-sub-styles-added!">🚀 V2 Update: Gender Overfit Fixed &amp; New Sub-Styles Added!</h3>Thank you to everyone who downloaded V1 and tipped Buzz! For V2, I completely rebuilt the model from scratch with a heavily upgraded dataset and expanded capabilities. I used 500 total images (adding 200 new images focused entirely on male subjects, non-humans, and items like armor/helmets) in attempt to break the female overfit. V2 now reliably handles any subject or gender you throw at it! (At least in my testing)What's New in V2:<ul><li>Sub-Style Triggers: While hand-refining the captions, I noticed the source style actually has a few distinct variations. I’ve trained V2 to recognize them! In addition to the base EvGaninStyle, you can now use:<ul><li>EvGaninFlatStyle - Angular, flat coloring with less detailed backgrounds.</li><li>EvGaninPainterlyStyle - Emphasizes paint-like strokes and dynamic line art thickness.</li><li>EvGaninComicStyle - A sharp, modern comic book aesthetic.</li><li>EvGaninCartoonStyle - Softer, cel-shaded looks with rounded, highly exaggerated proportions.</li></ul></li><li>Weight &amp; Usage: The sweet spot is still 1.0 to 3.0. Even at 1.0, the model is much stronger than just "inspired by." Pushing the weight higher will aggressively force the style and those signature exaggerated proportions. (And yes, just like V1, this is fully NSFW capable).</li><li>Training Note: I tested multiple epochs and settled on Epoch 20 for this release, as it had the absolute best grasp of the distinct sub-styles. If the community finds it a bit too strong or inflexible at higher weights, let me know in the comments and I can happily upload an earlier, slightly softer epoch as an alternative!</li></ul>A Humble Request for Support: Just like V1, I am releasing V2 100% free right out of the gate. Seeing the large number of downloads on V1 was incredible, but training 500 images at 3000 steps from scratch on the Civitai cloud trainer eats up a lot of Buzz! If you enjoy this model and want to help me continue releasing these massive Krea 2 LoRAs for free, please consider tossing a Buzz tip my way here on Civitai, or supporting my ongoing work over on my Patreon. Any support is massively appreciated and goes directly into cloud compute for the next release!<hr /><h3 id="v1-usage-strengths-and-limitations">⚠️ V1 Usage, Strengths, &amp; Limitations</h3>I am releasing this V1 so you can start experimenting, but please be aware of a specific quirk! Because the dataset happened to be heavily female skewed, the model highly favors female subjects at higher strengths.<ul><li>Recommended Strength: Works best between 1.0 and 3.0.</li><li>The Quirks (Gender Bias): Below 2.0, you get a great "inspired by" feel and the model obeys gender tags reasonably well. However, if you push the strength above 2.0, the style becomes much more pronounced, but the model will begin ignoring male/other prompts and heavily force a female character.</li><li>The Plan for V2: I will be doing a V2 retrain soon with a heavily balanced dataset to fix the gender bleeding. For now, V1 is fun and highly stylized, just keep an eye on your weights!</li></ul>Trigger Word: EvGaninStyle<h3 id="training-details-and-community-shoutout">🛠️ Training Details &amp; Community Shoutout</h3><ul><li>Dataset: ~300 curated images</li><li>Specs: 2000 steps, 20 Epochs, Rank/Dim/Alpha 16, LR 0.0001</li><li>Trainer: Civitai On-Site Cloud Trainer</li></ul>I want to give a massive shoutout to everyone who supported my previous concept models! Because you guys supported my temporary Early Access releases (which are now completely free!), I was able to use that Buzz to offload this massive ~300-image dataset to the Civitai cloud trainer. My local hardware struggles with anything over 100 images, so your support directly funded this model and saved my GPU!<h3 id="support-temporary-buzz-access-and-commissions">☕ Support, Temporary BUZZ Access, &amp; Commissions</h3>This Model is being released completely for free! But due to training costs I'm desperately pleading for any Buzz donations or other support if you are able!I still have a massive backlog of characters, concepts, and styles to train for Krea 2. If you want to support my work directly, you can tip Buzz, or check out my Patreon. I am also currently open for custom LoRA / LoKR training commissions (private use or expedited public release). Reach out via DM if you have a project in mind! (Note: I do not accept commissions to train on IRL persons).

Ev Ganin Style (Krea2 Style LoRA)

Full retrain. Worked out the higher strength female training bias for the most part.

Files with SHA256: 41375f0f125750615d284fb1ae61e7779b2d72d9563d974fb90cc1a5afb54fe2